I have an unbalanced textual dataset. Now when I perform K-Means clustering on that with the prior information that the dataset needs to be divided into two cluster(Elbow analysis and Shilloute Score), data points from one class gets distributed across 2 clusters. For example, Class A contains 500 Samples, Class B containing 300 Samples and Class C contains 400 Samples. When ran with K-Means Cluster 1 contains 300 samples of class A and 200 samples of class B. Hence I want to combine all the samples from class A to become one sample and so on. Now the problem here would be that I am not sure if I do this or not. I have done research and couldn't find anything of this sort. So I want to perform Observation Weighted K-Means, however I am not sure how do I calculate these weights?.
Any Help or suggestion is welcomed.